AI for Network Orchestration

Problem Context

In 2030 and beyond it is expected that businesses and consumers will demand for services that require even faster data rates and lower latencies than provided by current communication systems, and with virtually no service disruptions. Examples of these services are immersive/holographic communication and collaboration, the metaverse, and massive Internet-of-Things. All this is expected to be delivered on a massively flexible infrastructure capable to provide many different, and sometimes conflicting, features dependent on the need.

Solution

The solution developed within this project allows to train a Deep Reinforcement Learning (DRL) model for dynamic and intelligent network orchestration with resource allocation, user demand admission and routing in a resource-constrained edge environment. Specifically, we developed a network simulator and a set of integration scripts for training the DRL model, and a Proof of Concept for validating the model in a simulated network environment with an Event Uplink Streaming service. The DRL model is trained with the objective of maximizing user demand admission and Service Level Agreement (SLA) satisfaction, while minimizing resource consumption by performing service consolidation.

Results

This project introduces an autonomous AI system for service and resource management in Mobile Communication Networks, which is capable of self-adapting through interactions with the mobile network's open environment. This PoC implementation can also be re-purposed and re-used to evaluate and validate other decision models or algorithms for network resource allocation.